Most work is never written down

Ask someone who is good at their job to write down exactly how they do it. Pick something they do every week: preparing a quote, onboarding a new customer, closing the month, shipping a release. Most people can’t. They know how to do it, but that knowledge lives in their head and their hands. And if you watch them do it three times, you see three slightly different versions.

That is how most work runs. I think it is also one of the biggest opportunities in almost every company.

Talent, memory, and improvisation

In most organizations, things get done because someone is good at them. They remember the steps, notice when something is off, and adapt when the situation changes. Over time, each person finds their own way to repeat what they do. But that way stays with them. It isn’t written down, it isn’t measured, and it isn’t really repeatable. Each run is a little different from the last.

This works, and often it works well. Talent, experience, and the ability to adapt are real strengths. The trouble starts with a simple question: how well is this going?

Without a written process, there is nothing to compare against. Nobody knows how long the job really takes, how often it goes wrong, or what it costs. Nobody can say whether there is a better way. Sometimes people do feel it. They sense the job could be faster, cheaper, or better, but they can’t say how. They have never looked at the job as a process, only at the task in front of them.

Count the fires

Think about yesterday. How many small crises did you deal with? A customer got a different answer than last time. An order went out without the one detail that mattered. Someone did their part in a way nobody downstream expected, and you spent an hour putting it right. A deadline slipped because a step everyone assumed somebody else had done was never done.

Some of these are small. Some you catch just in time. A few turn into real problems a week later, when nobody remembers where they started. Most of us handle several of them every day, and we are good at it. So good that we stop seeing them. Putting out fires becomes part of the job, and the skill to do it becomes part of what we call talent.

Almost every one of these fires has the same root. Something was done in a way the next person didn’t expect, or it wasn’t thought through before it was handed on. That isn’t bad luck. It is what happens when work runs on memory.

Even in technical work

You might expect engineers and technicians to be the exception. Their fields are deterministic: the same input should give the same output. In my experience, they mostly work like everyone else. A release goes out by hand, a little differently each time. A server gets set up from memory. The one person who knows how the test environment works is on holiday, so nobody touches it.

Even where a procedure exists, it is usually applied roughly. A step gets skipped because it is usually fine. A checklist gets ticked from memory, after the work is done. If I had to guess, nine out of ten people never follow a written procedure for the work they do every week. I can’t prove that number. I just rarely see the opposite.

Human error is where the question starts

When something goes wrong, the usual explanation is human error. Someone forgot, misread, or was tired. It is accepted almost everywhere, even in work where mistakes are expensive or dangerous. And it ends the conversation: the person gets the blame, everyone promises to be more careful, and a few months later the same thing happens again, to someone else.

Safety research sees it the other way round. Sidney Dekker, whose books shaped much of modern safety thinking, puts it plainly: “Human error is not a cause of failure. Human error is the effect, or symptom, of deeper trouble.” If one tired person on a busy day can cause a failure, the way the work is set up allowed it. The useful question isn’t who made the mistake. It is why the work made the mistake easy, and what will catch it next time.

Source: Sidney Dekker, The Field Guide to Understanding Human Error.

What Cynefin says about it

The Cynefin framework helps me here. It describes four kinds of situations:

• Clear: cause and effect are obvious, and there is a right way to do things.
• Complicated: there is an answer, but finding it takes analysis or expertise.
• Complex: outcomes come from interactions you can’t work out in advance, so you try small things and learn.
• Chaotic: nothing holds, and you act first to make things stable.

Clear and complicated work are ordered, because cause and effect can be known. That is where a written process pays off. Yet most organizations treat much of their ordered work as if it were complex or chaotic. They improvise every time where a known way of doing it would do. The work feels chaotic, but much of it is only unwritten.

That doesn’t mean everything should become a procedure. Complex work needs room to experiment, and a crisis needs action before analysis. Work also doesn’t move neatly from one domain to the next. Untangling a job means sorting its parts: the steps that should be the same every time, the decisions that need an expert, and the parts where you still learn by trying. The aim is to stop improvising where it isn’t needed, so people’s talent goes where it is. More on Cynefin and what to automate first

Excellence is a process

The fields we trust most for precision got there through written, repeatable process.

On 30 October 1935, the Boeing Model 299, the prototype of the B-17, crashed on takeoff at Wright Field. Its elevator and rudder controls were still locked. Two of the men aboard, Major Ployer P. Hill and Boeing test pilot Leslie Tower, later died from their injuries. The crew was experienced, so telling pilots to be more careful would not have fixed it. The accident helped push the Army Air Corps to write checklists for takeoff, flight, landing, and after landing. Checklists existed before, but the crash helped make them standard as aircraft grew more complex.

Surgery is another field full of highly skilled people. A 2009 study across several hospitals introduced a 19-item surgical safety checklist. Afterwards, complications fell from 11.0 to 7.0 percent, and deaths in hospital from 1.5 to 0.8 percent. It was a before-and-after study, but the size of the change is hard to ignore.

In manufacturing, Taiichi Ohno, who shaped the Toyota Production System, put it in one line: “Where there is no standard, there can be no kaizen.” Kaizen means continuous improvement. Without a written way of doing something, you have nothing to improve on.

In all three cases the process didn’t replace the experts. It caught what experts miss on a busy day, and it gave them a baseline to improve on.

Sources: National Museum of the U.S. Air Force, Model 299 crash; Air & Space Forces Magazine, The checklist; Haynes et al., NEJM 2009; Selected sayings of Taiichi Ohno; Lean Enterprise Institute, Standardization is a countermeasure.

How to untangle a job

You can start this week, with one job:

1. Pick a job that repeats: weekly, monthly, or for every new customer.
2. Write down how it was actually done the last three times, rather than how it should be done.
3. Compare the runs. What stayed the same? What changed, and why?
4. For the parts that should be the same every time, agree on one way and write it as a short checklist.
5. Measure one or two things: how long it takes, how often it goes wrong, how much has to be redone.
6. Run it the same way a few times. When something goes wrong, change the checklist, not the person.
7. Once a part is clear and stable, hand it to an agent.

Where AI fits

This is the step before automation, and it is often the most valuable one. An agent needs a process. A talented person can improvise through a vague job; an improvising agent produces results you can’t predict. So the first thing I do in any project is write the work down with the people who do it.

AI also makes untangling cheaper than it used to be. It can turn a recorded walkthrough or a pile of old emails into a first draft of the process. It can compare how different people do the same job, and keep the checklist current when things change. Once the clear parts run through an agent, they run the same way every time, and every run leaves a record. For many teams, that is the first time they have real numbers about their own work.

Every job can be done well, every time

Precision and predictable quality aren’t reserved for airlines, hospitals, and car factories. They can come to every small piece of work: a quote, an onboarding call, a support reply, a monthly report, a release. With them come fewer fires. It starts with writing down what you already do.

If you have a job you’d like to untangle, write to me.